Abstract

As sensor investment decisions become increasingly crucial across industries, the challenge of optimal sensor selection has gained prominence. Traditional approaches primarily address this problem from a performance-oriented perspective, focusing on metrics like prediction or estimation accuracy. However, these methods often overlook the economic value that sensors can contribute. This paper addresses this gap by proposing a framework for evaluating sensors based on their economic contributions. Our approach leverages cooperative game theory and integrates explainable AI (XAI) techniques for feature valuation to assess the impact of each sensor within prediction models. We introduce two valuation methods—Marginal Influence Between Models (MIBM) and Marginal Influence Within Models (MIWM)—and provide guidelines on when to apply each method depending on the scenario. The MIWM method is found to be more effective in capturing the interactions and synergies among sensors, although it comes with higher computational costs. On the other hand, MIBM is advantageous in scenarios where computational efficiency and robustness are prioritized. To support method selection, we also analyze the computational complexity of both approaches and derive error bounds. Through numerical experiments using real-world semiconductor manufacturing data, we demonstrate that sensor selection based on economic value can lead to significantly different outcomes compared to conventional methods. Our findings show the importance of considering both economic benefit and cost factors in sensor selection, particularly in applications involving multiple sensor clusters or heterogeneous data sources.